openSAP

Introduction to Statistics for Data Science

Week 3 Unit 1

00:00:06 Hello, and welcome back to week three of the openSAP course An Introduction to Statistics for
Data Science.
00:00:15 In this unit we will look at correlation as a statistical measure.
00:00:21 When two sets of data are strongly linked together, we say they have a high correlation.
00:00:29 Correlation is positive when the values increase together, and correlation is negative when one
value
00:00:37 decreases as the other increases. Correlation can have values
00:00:43 ranging from minus one to plus one. One is a perfect positive correlation.
00:00:49 Zero is no correlation, the values don't seem to be linked at all.
00:00:54 And minus one is a perfect negative correlation. In this example, over the past two weeks,
00:01:02 a shop that sells ice creams has measured the temperature each day and the sales values
00:01:09 of the ice creams that were sold. This is plotted on a scatter plot shown in the slide.
00:01:16 You can see that there are higher sales values when the temperature increases,
00:01:22 so there is a correlation between sales and temperature. The Pearson product-moment
correlation coefficient
00:01:35 is a measure of the strength and direction of the linear relationship between two variables.
00:01:41 It's referred to as Pearson's correlation or simply as the correlation coefficient
00:01:48 and is calculated using the formula shown on the slide. If you are familiar with entering
functions
00:01:56 in Microsoft Excel you could enter the CORREL command, and that will allow you to calculate
00:02:04 the relationship between the variables. If the relationship is non-linear,
00:02:09 then the correlation coefficient does not adequately represent the strength
00:02:15 of the relationship between the variables. In the ice cream example,
00:02:22 temperature is x and ice cream sales is y. Follow these steps to calculate
00:02:29 Pearson's correlation coefficient. The first step is to find the mean of x, and the mean of y.
00:02:36 And then you can subtract the means from every value and follow other steps through
00:02:42 in this calculation that's actually showing on the slide for you.
00:02:49 Finally, you divide all the sums, and this will allow you to calculate the Pearson's correlation
coefficient,
00:02:58 which in this example is +0.96, which indicates a very high positive correlation
00:03:04 between temperature and ice cream sales. Pearson's correlation is only relevant
00:03:12 when the relationship is linear. If the data in the example is extended with points
00:03:17 that reflect that sales reduce if temperatures increase further,
00:03:23 then the scatter plot could be as it's shown in the slide. In this example, customers will stop
buying ice creams,
00:03:31 maybe because it's so hot they don't go shopping. The correlation coefficient for this data is
zero,
00:03:40 indicating there is no correlation. However, the scatter plot shows there is a correlation,
00:03:47 but it's a curve, peaking at around 25 degrees Centigrade. You now understand why it's
important to visualize
00:03:55 the data when you are analyzing correlations. These are the scatter plots of Anscombe's
quartet.
00:04:04 It's a set of four different pairs of variables created by the English statistician Frank Anscombe
in 1973.
00:04:14 These graphs were constructed to demonstrate both the importance of graphing data before
analyzing it
00:04:23 and the effect of outliers on statistical properties. The first graph, on the top left,
00:04:30 seems to be distributed normally, and the two variables appear to be correlated.
00:04:37 The second graph, on the top right, is not distributed normally. There is an obvious relationship
between the two variables,
00:04:45 but it's not linear. In the third graph, the bottom left,
00:04:50 the linear relationship is perfect, except of course for one outlier
00:04:55 which exerts enough influence to lower the correlation coefficient from one down to 0.816.
00:05:04 In the fourth graph, an outlier produces a high correlation coefficient,
00:05:09 even though the relationship between the two variables is non-linear.
00:05:15 However, the four y variables have the same mean, variance, and correlation, although the
distribution
00:05:22 of the variables is very different. These examples indicate that the correlation coefficient,
00:05:30 as a summary statistic, cannot replace visual examination of the data.
00:05:36 You should visualize the data and calculate the correlation coefficient.
00:05:43 So to summarize: Correlation measures the strength
00:05:47 of association between two variables. The most common correlation coefficient is called
00:05:54 the Pearson product-moment correlation coefficient. The sign and the absolute value of a
Pearson correlation
00:06:04 coefficient describe the direction and the magnitude of the relationship between two variables.

00:06:13 The value of a correlation coefficient ranges between minus one and plus one.
00:06:18 The greater the absolute value of a correlation coefficient, the stronger the linear relationship.
00:06:27 The strongest linear relationship is indicated by a correlation coefficient of minus one or plus
one.
00:06:34 The weakest linear relationship is indicated by a correlation coefficient of zero.
00:06:40 A positive correlation indicates that as one variable increases in value,
00:06:46 the other variable tends to increase as well. A negative correlation indicates
00:06:52 that as one variable increases in value, the other variable tends to reduce in value.
00:06:59 Remember that the Pearson correlation coefficient only measures linear relationships.
00:07:07 A correlation of zero does not mean zero relationship between the two variables;
00:07:13 it means zero linear relationship. It's possible for two variables to have zero linear
00:07:20 relationship and a strong curvilinear relationship at the same time.
00:07:26 So always remember to use scatter plots to visualize the relationship as well
00:07:32 as calculating the correlation coefficient. In the next unit, we will look
00:07:38 at correlation versus causation.

2
Week 3 Unit 2

00:00:06 Hello, and welcome back to week three, unit two, where we're going to consider correlation
versus causation.
00:00:15 You've learned in a previous unit that correlation is not causation.
00:00:21 If two variables are highly correlated, it does not mean that one thing causes the other.
00:00:29 There can be many reasons why variables are correlated. However, you do need to think
about
00:00:36 the relationship carefully because one variable might cause the other or might not.
00:00:42 If the ice cream shop measures the daily sales of sunglasses sold by an optician
00:00:48 and compares them to daily ice cream sales, then there is a clear high, positive correlation.
00:00:56 But sunglasses do not cause people to buy ice cream. Numerous epidemiological studies
showed
00:01:05 that women taking combined hormone replacement therapy, HRT, also had a lower than
average incidence
00:01:14 of coronary heart disease, CHD, leading doctors to propose that HRT
00:01:19 was in some way protective against heart disease. However, randomized controlled trials
showed
00:01:25 that HRT caused, in fact, a small, but statistically significant increase
00:01:30 in the risk of coronary heart disease. Reanalysis of the data showed that women undertaking
HRT
00:01:37 were more likely to be from higher socio- economic groups, ABC1 groups, with better than
average diet
00:01:45 and exercise regimens. The use of HRT
00:01:48 and decreased incidence of coronary heart disease were coincident effects of a common
cause:
00:01:56 that is, the benefits associated with a higher socio-economic status,
00:02:01 rather than a direct cause and effect, as had been supposed. If two events, A
00:02:09 and B, are correlated, then the relationship between them could be expressed as follows:
00:02:15 A causes B, a direct causation; B causes A, reverse causation;
00:02:21 A and B are consequences of a common cause, but do not cause each other; A causes B
00:02:27 and B causes A, bidirectional causation; A causes C, which causes B, indirect causation;
00:02:35 or, in fact, there is no connection between A and B, and the correlation is a coincidence.
00:02:41 Therefore, knowing that A and B are correlated does not enable you to confirm
00:02:47 the existence or direction of a cause and effect. The response variable is the focus
00:02:56 of a question in a study or experiment. An explanatory variable is one that explains
00:03:04 changes in the response variable. If you suspect that X causes Y,
00:03:10 then there could be a spurious relationship if, in reality, X and Y are both caused by Z.
00:03:17 For example, when examining a city's ice cream sales, it seems that sales are higher when the
rate of drowning
00:03:25 in the city swimming pools is highest. However, to say that ice cream sales cause drowning,
00:03:33 or vice versa, implies a spurious relationship between those two variables.
00:03:39 In reality, a heatwave may have caused both. The heatwave is an example of a lurking
variable.
00:03:47 A lurking variable is one whose effects on the response variable cannot be distinguished
00:03:53 from one or more of the explanatory variables in the study, and it's not considered in the
design of the study.

3
00:04:03 A confounding variable is one whose effects on the response variable cannot be distinguished

00:04:09 from one or more of the explanatory variables in the study. The difference between lurking
00:04:18 and confounding variables lies in their inclusion in the study.
00:04:25 In experimental research, spurious relationships can often be identified by controlling for other
factors,
00:04:33 including those that have been theoretically identified as possible confounding factors.
00:04:41 For example, consider a scientist trying to determine whether a new drug kills bacteria.
00:04:48 When the drug is applied to a bacterial culture, the bacteria die.
00:04:54 To help rule out the presence of a confounding variable, another culture is subjected to
conditions
00:05:01 that are identical to those facing the first culture. However, the drug is not applied to the
second culture.
00:05:10 This is called the control. If there is an unseen confounding factor,
00:05:16 the control culture will also die. However, if the control culture does not die,
00:05:22 then the new drug is responsible for killing the bacteria. To summarize: Correlation is a
statistical measure
00:05:33 expressed as a number that describes the size and direction of a linear relationship
00:05:40 between two variables. Causation indicates that one event is the result
00:05:46 of the occurrence of another: In other words, there is a causal relationship between the two
events.
00:05:54 This is also referred to as cause and effect. For A to cause B, we tend to say that, at a
minimum,
00:06:02 A must precede B, the two must covary, vary together, and no competing explanation
00:06:09 can better explain the covariance of A and B. Taken alone, however, these three requirements

00:06:16 cannot prove cause. They are necessary, but they're not sufficient.
00:06:22 Lurking and confounding variables can make it difficult to conclude that it was the explanatory
variables alone
00:06:32 that affected the observed changes in the response variable. In the next unit, we will look at
scatter plots
00:06:40 and line of best fit.

4
Week 3 Unit 3

00:00:06 Hello, and welcome back to week three, unit three, where we're going to consider scatter plots

00:00:12 and the line of best fit. You've seen in previous lessons
00:00:19 that you can observe trends in data by creating a scatter plot,
00:00:24 which is a two-dimensional graph of y versus x, where x and y are two of the quantities in your
data set.
00:00:34 What you hope to see in a scatter plot is the relationship between x and y,
00:00:39 in which the value of quantity y, which is called the target or dependent variable,
00:00:46 in some way, depends on the value of quantity x, which is called the explanatory or
independent variable.
00:00:55 This example plot indicates there's an underlying relationship between x and y.
00:01:04 A scatter plot can be used to show the relationship between two variables.
00:01:09 So in this example, the underlying trend in the data is a straight line,
00:01:15 so the relationship between x and y is linear. You could eyeball the graph,
00:01:21 and with a pencil and a straightedge or ruler, sketch in the line where it appears to fit the data,

00:01:29 estimate its slope and the y-intercept, and be satisfied that your estimate
00:01:35 is close enough to meet your needs. But if you are looking for something
00:01:40 a little bit more accurate, you can obtain the mathematical definition
00:01:45 of the line that most accurately represents the data. This is called line of best fit.
00:01:54 The line of best fit is the best possible straight line that fits the data.
00:02:00 Sometimes the line of best fit is referred to as the trend line.
00:02:05 It's the line where the sum of the squares of the residuals, which are the errors between the
individual data values
00:02:14 and the line, is at its minimum. The slope and the y-intercept are two numbers
00:02:21 needed to define the equation of a line, which is y equals mx plus b.
00:02:26 So there are two formulas that you need to define the line of best fit,
00:02:31 for the slope, m, and the intercept, b, and these, of course, are given on the slide.
00:02:37 So let's return to the temperature and ice cream example. The scatter plot enables you to
visualize the relationship.
00:02:48 The first step is to calculate the mean of the x, which is x bar,
00:02:52 by adding up all of the x values and dividing by n, the number of observations.
00:02:58 Step two is to calculate the mean of y, which is called y bar.
00:03:02 Again, it's calculated just like you did for x. Step three is to subtract the mean of x
00:03:10 from every x value, and the mean of y from every y value.
00:03:15 Step four is to calculate x minus x bar times y minus y bar, which is the penultimate column
00:03:21 in the table, and x minus x bar squared, the last column in the table,
00:03:27 for each of the different values. Step five is to calculate the slope.
00:03:32 And step six is to calculate the y-intercept. So this will give you the equation
00:03:38 that you're seeing, which is the line of best fit. The trend line and equation of the line
00:03:45 can be added to the scatter plot in Microsoft Excel, as you can see here.
00:03:51 So to summarize: A scatter plot can be used
00:03:55 to show the relationship between two variables. The line of best fit is the line
00:04:01 that describes that relationship between the two variables,
00:04:04 where the sum of the squares of the residual errors between the individual data values

5
00:04:12 and the line is at its minimum. Therefore, it's the best straight line that fits the data.
00:04:20 The slope, m, and the y-intercept, b, are the two values that are needed to define
00:04:27 the equation of a straight line, y equals mx plus b. So in the next unit, we're going to consider
linear regression.

6
Week 3 Unit 4

00:00:06 Hello, and welcome back to week three, unit four, where we're going to be considering linear
regression.
00:00:15 Linear regression is an approach to modeling the linear relationship
00:00:19 between a target variable, also referred to as the dependent variable,
00:00:25 and one or more explanatory variables, known as independent variables.
00:00:30 The case of one explanatory variable is called simple linear regression.
00:00:37 For more than one explanatory variable, the process is called multiple linear regression.
00:00:46 Most applications of linear regression fall into one of the following two broad categories.
00:00:54 Firstly, prediction or forecasting. Linear regression can be used to fit a predictive model
00:01:01 to an observed set of data values of the target and explanatory variables.
00:01:08 After developing the model, when additional values of the explanatory variables
00:01:13 are collected without an accompanying target value, then the model can be used to make a
prediction
00:01:21 of the target values. Secondly, explaining variation in the target variables
00:01:27 that can be attributed to variation in the explanatory variables.
00:01:35 Linear regression analysis can be applied to quantify the strength of the relationship
00:01:40 between the targets and the explanatory variables. It can be used to determine if some
explanatory variables
00:01:50 have no linear relationship with the target variable. Linear regression models are often
00:02:00 fitted using the least squares approach, although they may also be fitted in other ways.
00:02:09 The best fit from a least squares perspective is to minimize the sum of squared residuals,
00:02:16 when a residual is the difference between an observed value, and the fitted value provided by
a model:
00:02:24 in other words, the error. In the ice cream example,
00:02:28 we can calculate the residual, and this is shown in the table.
00:02:33 Ordinary least squares OLS is a type of linear least squares method
00:02:38 for estimating the unknown parameters in a linear regression model.
00:02:44 OLS chooses the parameters of a linear function of a set of explanatory variables
00:02:50 by the principal of least squares. That's minimizing the sum of the squares of the differences
00:02:58 between the observed target or dependent variables, that is, the values of the variables being
predicted,
00:03:06 and those predicted by the linear function itself. We can use linear regression to forecast target
values.
00:03:17 The relationship between the target and explanatory variables
00:03:20 is modeled using linear predictive functions. The unknown model parameters in the regression
equation
00:03:29 are estimated from the data. In the example, the linear regression model
00:03:36 is sales in dollars equals 29.82, temperature and degrees, 145.71.
00:03:42 Therefore, if the temperature tomorrow is forecast to be 20 degrees,
00:03:46 then the store can expect to sell 20.82 times 20 is 145.71. Over $450 worth of ice cream.
00:03:56 Least squares works by making the total of the square of the residuals as small as possible.
00:04:04 The straight line – that is, the regression line – minimizes the sum of squared residuals.
00:04:11 In the graph on the right, you can also see how an unusually high or low value,
00:04:16 an outlier, will have a high influence on the model, as it will pull the regression line towards it.
00:04:28 Standard linear regression models with standard estimation techniques

7
00:04:33 make a number of assumptions about the predictor variables, the response variables, and their
relationship.
00:04:44 The main assumptions are usually listed as follows. Firstly, there is a linear relationship
00:04:49 between the explanatory and target variables. Secondly, there is no or low multicollinearity.
00:04:57 This is correlation between the explanatory variables. There's also no auto-correlation.
00:05:04 This occurs when the residuals are not independent from each other.
00:05:09 In other words, when the value of y(x+1) is not independent from the value of y(x),
00:05:18 the data is homoscedastic, meaning the residuals are equal across the regression line.
00:05:24 Numerous extensions and advances to the basic regression approach
00:05:29 have been developed that allow each of these assumptions to be relaxed – in other words,
reduced to a weaker form –
00:05:37 and in some cases, eliminated entirely. Please note that there are many articles
00:05:44 relating to this subject on the internet. And some list a number of additional assumptions.
00:05:54 Linear regression requires the relationship between the explanatory and target variables
00:05:59 to be linear. This assumption addresses the functional form
00:06:05 of the model. The regression model is linear when all terms
00:06:10 in the model are either the constant or a parameter multiplied by an explanatory variable.
00:06:18 You build the model equation only by adding the terms together.
00:06:23 This rule constrains the model to one type shown in the slide.
00:06:29 In the equation, Y is the target variable you're trying to predict.
00:06:35 The X, X1 to Xk, are the explanatory variables you're using
00:06:39 to predict the target. The betas, β1 to βk are the coefficients
00:06:44 or multipliers that describe the size of the effect the explanatory variables are having on the
target Y.
00:06:52 These are the parameters that ordinary least squares process estimates.
00:06:58 β0 is the value Y is predicted to have when all the explanatory variables are equal to zero,
00:07:05 the Y intercept. Epsilon is the random error.
00:07:09 The linearity assumption can be tested with scatter plots. You can see there are three simple
examples
00:07:17 showing a strong linear relationship, a weaker linear relationship
00:07:21 and no linear relationship at all between the target Y variable
00:07:25 and the explanatory X variable. Also please be aware that linear models
00:07:32 can model curvature by including non-linear explanatory variables,
00:07:38 such as polynomials and exponential functions. Multicollinearity occurs when the explanatory
variables
00:07:49 are highly correlated with each other. If multicollinearity is found in the data,
00:07:57 the simplest way to address the problem is to remove one of the correlated variables.
00:08:03 However, there are a range of more sophisticated techniques available.
00:08:13 Linear regression analysis requires that there is little or no autocorrelation in the residuals.
00:08:20 This means that the error terms must be uncorrelated so that one observation of the error term

00:08:27 should not predict the next. Autocorrelation occurs when the error terms
00:08:34 are not independent from each other. In other words, when the value of y(x+1)
00:08:41 is not independent from the value of y(x). For instance, if the error for one observation is
positive,
00:08:50 and that increases the probability that the following error is positive,
00:08:55 then there is a positive correlation. You can assess if this assumption is violated

8
00:09:02 by graphing the residuals in the order that the data were collected,
00:09:07 you hope to see a randomness in the plot. While a scatter plot allows you
00:09:13 to check for autocorrelation, you can also test the linear regression model
00:09:18 for autocorrelation with the Derbin Watson test. The variance of the errors should be
consistent
00:09:29 for all observations. This means that the variance does not change
00:09:34 for each observation or for a range of observations. The scatter plot is a good way to check
00:09:43 whether the data are homoscedastic, which simply means that the residuals are equal
00:09:49 across the regression. You can check this assumption by plotting the residuals
00:09:56 against the fitted values. Heteroscedastic appears as a cone shape,
00:10:02 where the spread of the residuals increases in one direction.
00:10:06 The scatter plot on the right shows an example of data, though not homoscedastic,
00:10:13 they're heteroscedastic. Please also note that when assumption three,
00:10:18 no auto correlation that is, and four, homoscedastic, are both true,
00:10:22 statisticians say that the error term is independent and identically distributed.
00:10:29 They refer to this as spherical errors. Although there was no assumption about the distribution

00:10:38 of the explanatory variables, it is good practice to examine and understand the data
00:10:45 before you build a model. You can check the distribution visually,
00:10:50 for example with a histogram, a box plot, or a Q-Q plot. A Q-Q, Quantile-Quantile, plot is a plot
00:10:58 of the quantiles of two distributions against each other, or a plot based on estimates of the
quantiles.
00:11:06 The pattern of points in the plot is used to compare two distributions.
00:11:11 So to test for normality, you could use it to compare the distribution
00:11:15 of an explanatory variable to the theoretical normal distribution.
00:11:22 If the two distributions being compared are similar, the points in the Q-Q plot
00:11:28 will approximately lie on the line y = x. However, if the distributions are linearly related,
00:11:36 the points in the plot will approximately lie on a line. But not necessarily on the line y = x.
00:11:45 To summarize: Linear regression is an approach to modeling the linear relationship
00:11:51 between a target variable and one or more explanatory variables.
00:11:58 Simple linear regression has one explanatory variable, and multiple linear regression
00:12:03 has more than one explanatory variable. In summary, your linear regression model
00:12:11 should produce residuals that have a mean of zero, have a constant variance and are not
correlated
00:12:19 with themselves or other variables. If these assumptions are true,
00:12:25 then the ordinary least squares regression procedure will create the best possible estimates.
00:12:34 In the next unit, we will consider how to interpret the results from a regression.

9
Week 3 Unit 5

00:00:06 Hello, and welcome back to the fifth and final unit of week three, where we're going to consider

00:00:13 how to interpret the results of a linear regression. Regression analysis is used to produce an
equation
00:00:21 that will predict a target variable using one or more explanatory variables.
00:00:28 This equation has the form shown in the slide, where Y is the target variable you are trying to
predict.
00:00:37 The Xs, X1 to Xk, are the explanatory variables you are using to predict the target.
00:00:44 And the betas, β1 to βk are the coefficients that describe the size of the effect
00:00:50 the explanatory variables are having on the target variable Y.
00:00:55 These are the parameters that the ordinary least squares process estimates.
00:01:01 β0 is the value Y is predicted to have when all the explanatory variables are equal to zero,
00:01:08 it's the Y intercept. Epsilon is the random error.
00:01:15 This is a simple example using the R package in RStudio. You can install this free-to-use
software if you want to
00:01:24 and explore the tremendous functionality that's available. There are many user guides you can
download,
00:01:33 so you can start to use the package. We'll make the data available for you.
00:01:40 The data shows the price obtained for a number of properties.
00:01:45 This is called the target variable or the dependent variable.
00:01:50 The data also includes the number of bedrooms, bathrooms, the size of the living area in
square feet,
00:01:58 and of the overall lot size, and the number of floors in the house.
00:02:03 These are the explanatory variables or the independent variables.
00:02:07 Initially, you could run a summary to look at the range, the quartiles, median, and mean of the
variables.
00:02:16 This example data has been provided for you if you would like to repeat this exercise.
00:02:22 There are links to download RStudio in the slide. The data frame is called house_train.
00:02:31 The data summary uses the R command "summary". The number of rows and columns
00:02:37 is given by the R command "dim". This initial analysis can often indicate
00:02:43 interesting features in the data. In this example, there is a similarity in the data summary
00:02:49 for the explanatory variables "bedrooms" and "bathrooms", and this will require further
investigation.
00:03:01 You might also want to visualize the data. This is a simple visualization using the R command
plot.
00:03:09 Here you can see the basic relationship between each of the variables in the data.
00:03:16 You will notice that there is a strong correlation between the number of bedrooms and
bathrooms.
00:03:26 This might indicate that there is a problem in the data, and this should trigger some further
analysis
00:03:34 so that you can discover what's happening in the data. You can fit a multiple linear regression
model
00:03:42 with price as the target variable, and the other variables as the explanatory variables.
00:03:50 The R command is given here in the slide. This is the output
00:03:56 when you build the linear regression model. The coefficient terms can be represented
00:04:03 in the equation shown in the slide. Note that there is no coefficient for bathrooms X2,

10
00:04:12 you'll explore why this is later in this lesson. You will need to test
00:04:18 whether the explanatory variables in the model collectively have an effect on the target
variable.
00:04:26 To do this, you test what is called the null hypothesis, which is represented as H0,
00:04:33 against what is called the alternative hypothesis H1. This is a simple test.
00:04:40 If there is a significant linear relationship between the explanatory variables, the Xs,
00:04:47 and the target variable Y, then the slope will not equal zero.
00:04:52 The null hypothesis states that the slope is equal to zero, and the alternative hypothesis states
00:04:59 that the slope is not equal to zero. When you test the null hypothesis
00:05:05 it simply means you are assessing the probability that there is no relationship
00:05:11 between the explanatory variables and the target variable. You then either accept or reject this
null hypothesis.
00:05:21 Therefore, the hypotheses read like this: H0 is where beta one equals beta two equals zero.
00:05:29 And H1 is where at least one beta value is not zero. The F-statistic is the test statistic
00:05:37 used to decide whether the model as a whole has statistically significant predictive capability.
00:05:44 The F-statistic is the ratio of the mean regression sum of squares
00:05:49 divided by the mean error sum of squares. Its value will range from zero up to a large number.

00:05:58 And the value represents the probability that the null hypothesis for the full model is true:
00:06:04 in other words, that all of the regression coefficients are zero. In general, if your calculated F-
statistic
00:06:12 of your regression model is larger than the threshold critical F-statistic
00:06:18 given by F-distribution tables or calculated in your statistical software,
00:06:25 you can reject the null hypothesis. The p-value is determined by the F-statistic
00:06:32 and is the probability your results could have happened by chance.
00:06:37 If the p-value is less than the alpha level, commonly set at 0.05, you can reject the null
hypothesis.
00:06:46 In our example, the output shows that F equals with a very small P value,
00:06:53 and that indicates you can reject the null hypothesis that the explanatory variables collectively
00:06:59 have no effect on the target variable. If you can reject this null hypothesis,
00:07:07 you continue by testing whether the individual regression coefficients are significant
00:07:14 while controlling for the other variables in the model. The p-value for each individual variable
00:07:22 tests the null hypothesis that the coefficient is equal to zero:
00:07:27 that is, there is no effect. A low p-value less than 0.05 indicates
00:07:33 that you can reject the null hypothesis. So basically, the explanatory variable
00:07:39 that has a low p-value is likely to be a meaningful addition to your model
00:07:45 because changes in the variable's value are related to changes in the target variable.
00:07:51 Conversely, a larger insignificant p-value suggests that changes in the explanatory variable
00:07:59 are not associated with changes in the target. You use the coefficient p-values
00:08:06 to determine which terms to keep in the regression model. The standard errors of the
coefficients
00:08:16 are the estimated standard deviations of the errors in estimating them.
00:08:22 The larger the standard error of the coefficient estimate, the less precise the measurement of
the coefficient.
00:08:31 The t value is the coefficient estimate divided by its standard error.
00:08:38 In the output shown in the slide, the first column gives the estimated value
00:08:44 for each coefficient. The second column gives the standard error,

11
00:08:49 and the third column gives the t value. The output compares the t-statistic of the variable
00:08:57 with values in the student's t-distribution to determine the p-value.
00:09:02 The student's t-distribution describes how the mean of a sample with a certain number of
observations
00:09:10 is expected to behave. R-squared is a statistical measure
00:09:16 of how close the data are to the fitted regression. It's also known as the coefficient of
determination
00:09:24 or the coefficient of multiple determination for multiple regression.
00:09:29 R-squared is the percentage of the target variable variation that is explained by a linear model.

00:09:38 It's equal to the explained variation divided by the total variation.
00:09:43 It's always between 0 and 100% if there's an intercept value.
00:09:49 0% indicates that the model explains none of the variability of the target data around its mean.

00:09:56 100% indicates that the model explains all the variability of the target data around its mean.
00:10:04 However, there are two problems with R- squared. Firstly, every time you add an explanatory
variable
00:10:11 to a model, the R-squared increases, even if due to chance alone.
00:10:16 It never decreases. Therefore, a model with more terms
00:10:21 may appear to have a better fit simply because it has more terms.
00:10:26 Secondly, if a model has too many explanatory variables, it begins to model the random noise
in the data.
00:10:34 This is known as overfitting the model, and it produces misleadingly high R-squared values,
00:10:41 but the model is basically unable to make accurate predictions.
00:10:47 The adjusted R-squared is a modified version of R-squared that has been adjusted
00:10:53 for the number of explanatory variables in the model. The adjusted R-squared increases only
00:11:02 if the new term improves the model more than would be expected by chance.
00:11:08 It decreases when an explanatory variable improves the model by less than expected by
chance.
00:11:16 Note that the adjusted R-squared can be negative, and it is always lower than the R-squared.
00:11:26 Regression coefficients represent the mean change in the target variable for one unit of
change
00:11:32 in the explanatory variable, while holding other predictors in the model constant.
00:11:39 This statistical control that regression provides is important because it isolates the role
00:11:47 of one variable from all of the others in the model. For example, the equation shows
00:11:53 that the coefficient for the explanatory variable square foot of living is 3.281e squared.
00:12:01 The coefficient indicates that for every additional square foot in living space,
00:12:07 you can expect the target price to increase by an average of 3.281e squared dollars.
00:12:16 You may have noticed that there is an error identified in the output:
00:12:21 One coefficient is not defined because of singularities. This occurs because two of the
explanatory variables
00:12:29 are perfectly correlated: bedrooms and bathrooms. So the coefficient cannot be specified.
00:12:37 You can see this when you produce the correlation matrix using the R command given here.
00:12:46 Therefore, the explanatory variable "bathrooms" is omitted from the model.
00:12:54 So to summarize: Multiple linear regression is used to describe data
00:13:01 and to explain the relationship between one target variable
00:13:05 and two or more explanatory variables. The analysis requires you to analyze
00:13:13 the correlation and directionality of the data, train the model, and then evaluate

12
00:13:20 the validity and usefulness of the model. This unit introduced you to some of the results
00:13:27 that are generated that will help you evaluate your model. You must assess these very
carefully
00:13:34 so that you can be sure that your model is valid. Remember that it's also very important
00:13:42 that you check and confirm that the basic assumptions for linear regression
00:13:48 that were discussed in the previous unit hold true. With this, I'd like to close the third week.
00:13:56 I hope you enjoyed these units of this course, and we are happy to get in touch with you
00:14:02 in our discussion forum if you have any content-related questions.
00:14:08 Now, we will wish you all the best for the weekly assignment and see you next week,
00:14:14 where we will introduce probability and Bayes' Theorem.

13
www.sap.com/contactsap

© 2019 SAP SE or an SAP affiliate company. All rights reserved.

No part of this publication may be reproduced or transmitted in any form or for any purpose without the express permission of SAP SE or an SAP affiliate company.

The information contained herein may be changed without prior notice. Some software products marketed by SAP SE and its distributors contain proprietary software components of other software vendors.
National product specifications may vary.

These materials are provided by SAP SE or an SAP affiliate company for informational purposes only, without representation or warranty of any kind, and SAP or its affiliated companies shall not be liable
for errors or omissions with respect to the materials. The only warranties for SAP or SAP affiliate company products and services are those that are set forth in the express warranty statements
accompanying such products and services, if any. Nothing herein should be construed as constituting an additional warranty.

In particular, SAP SE or its affiliated companies have no obligation to pursue any course of business outlined in this document or any related presentatio n, or to develop or release any functionality
mentioned therein. This document, or any related presentation, and SAP SE’s or its affiliated companies’ strategy and possible future developments, products, and/or platform directions and functionality are
all subject to change and may be changed by SAP SE or its affiliated companies at any time for any reason without notice. The information in this document is not a commitment, promise, or legal obligation
to deliver any material, code, or functionality. All forward-looking statements are subject to various risks and uncertainties that could cause actual results to differ materially from expectations. Readers are
cautioned not to place undue reliance on these forward-looking statements, and they should not be relied upon in making purchasing decisions.

SAP and other SAP products and services mentioned herein as well as their respective logos are trademarks or registered trademarks of SAP SE (or an SAP affiliate company) in Germany and other
countries. All other product and service names mentioned are the trademarks of their respective companies. See www.sap.com/copyright for additional trademark information and notices.